Research Assistant

Turns rough research notes into illustrated ones. It finds every Bear note tagged 待整理 ("to be organized"), derives a topic from each note's content, searches for a relevant GIF, and inserts it into the note. Built on the bear-notes (grizzly) and gifgrep skills.

Platform: macOS only. Requires the Bear app running, a Bear API token, and the grizzly + gifgrep CLIs installed (see bear-notes and gifgrep skills).

Prerequisites

• Bear app installed and running.
• Bear API token saved at ~/.config/grizzly/token (Bear → Help → API Token → Copy Token).
• grizzly CLI (see bear-notes skill) and gifgrep CLI (see gifgrep skill).

Quick Start

# Dry run: show what would change, insert nothing
./research_assistant.sh --tag 待整理 --dry-run

# Real run: insert one matched GIF per note
./research_assistant.sh --tag 待整理

# Limit how many notes to process in one pass
./research_assistant.sh --tag 待整理 --max 5

How It Works

1. Collect notes — grizzly open-tag --name "待整理" --enable-callback --json lists every note carrying the tag.
2. Read each note — grizzly open-note --id <ID> --enable-callback --json pulls the full note body.
3. Derive a topic — the script takes the note title plus the most frequent meaningful words from the body to build a short GIF search query. This is intentionally simple and dependency-free; tune topic_query() if you want smarter extraction.
4. Match a GIF — gifgrep "<topic>" returns a relevant GIF URL.
5. Insert inline — the GIF is appended to the note as Markdown (![topic](gif-url)) via grizzly add-text --id <ID> --mode append --token-file ~/.config/grizzly/token. Use --mode prepend if you'd rather place it at the top.

Options

• --tag <name> — tag to scan (default: 待整理).
• --max <n> — process at most N notes (default: all).
• --mode <append|prepend> — where to insert the GIF (default: append).
• --dry-run — print planned actions without modifying any note.
• --token-file <path> — Bear token path (default: ~/.config/grizzly/token).

Notes & Limitations

• Idempotency: the script skips a note if it already contains a <!-- research-assistant:gif --> marker, so re-runs won't stack duplicate GIFs.
• Topic extraction is keyword-frequency based, not semantic. For niche notes, review the dry-run output before a real run.
• One GIF per note per pass by design — keeps notes readable.
• Bear must stay open and focused enough to honor x-callback-url requests.